基于4DVAR和EnKF的遥感信息与作物模型冬小麦估产  

Winter Wheat Yield Estimation Based on Assimilated Remote Sensing Date with Crop Growth Model Using 4DVAR and EnKF

作　　者：刘正春[1] 徐占军[1] 毕如田[1] 王超[2] 贺鹏[1] 杨武德[2] LIU Zhengchun;XU Zhanjun;BI Rutian;WANG Chao;HE Peng;YANG Wude(College of Resource and Environment,Shanxi Agricultural University,Taigu 030801,China;College of Agriculture,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学资源环境学院,太谷030801 [2]山西农业大学农学院,太谷030801

出　　处：《农业机械学报》2021年第6期223-231,共9页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家重点研发计划项目(2018YFD020040103)。

摘　　要：为了提高遥感信息与作物模型同化的估产精度,以山西省晋南地区的3个县为研究区,采用四维变分(Four-dimensional variational,4DVAR)和集合卡尔曼滤波(Ensemble Kalman filter,EnKF)两种同化算法将高时空分辨率Sentinel多源数据反演的叶面积指数(Leaf area index,LAI)、土壤含水率(θ)和CERESWheat模型进行同化,对比两种算法同化LAI和θ的性能,并进行冬小麦产量估测。结果表明:两种同化算法均能结合遥感观测和作物模型模拟的优势,相比模型模拟值,同化精度均有所提高;4DVARLAI和4DVARθ的均方根误差(Root mean square error,RMSE)分别比EnKFLAI和EnKFθ低0.1490 m2/m2、0.0091 cm3/cm3,且根据遥感实际监测值4DVARLAI更能精确识别冬小麦的物候期,与实际冬小麦生长发育的物候期更相符,因此在Sentinel多源数据与CERESWheat模型同化中,4DVAR算法的性能更好;由4DVAR同化后的LAI和θ双变量建立的估产模型,RMSE和平均相对误差(Mean relative error,MRE)小于CERESWheat模型模拟估产的RMSE和MRE,说明估产模型的估产误差小,采用4DVAR算法同化Sentinel多源数据与CERESWheat模型有效提高了冬小麦区域估产精度。To improve the precision of crop yield estimation by integrating the remote sensing data into the crop model,two methods were applied,the four-dimensional variational(4DVAR)and the ensemble Kalman filter(EnKF),to assimilate the leaf area index(LAI)and the soil moisture(θ)derived from Sentinel multi-source data with the CERESWheat model.The two algorithms were assessed on the performance of assimilation of LAI andθand estimated the yield of winter wheat across three counties located in the south of Shanxi Province in China.It was found that both assimilation algorithms can combine the advantages of remote sensing observations and crop model simulations.Compared with the crop model simulation values,the accuracy of assimilated LAI andθwere improved.Compared with EnKF,the 4DVAR algorithm can reduce the RMSEs of the assimilated LAI andθby 0.1490 m2/m2 and 0.0091 cm3/cm3,respectively.And 4DVARLAI could accurately identify the phenological period of winter wheat according to the remote sensing observations,which was more consistent with the growth and development of the actual phenological period of winter wheat.Therefore,4DVAR showed a better performance in the assimilation of Sentinel multi-source data with CERESWheat model.The accuracy of the yield estimation model based on assimilated LAI andθby 4DVAR(RMSE was 449.77 kg/hm2,MRE was 7.85%)was higher than the yield accuracy based on simulated values by the CERESWheat model(RMSE was 641.55 kg/hm2,MRE was 10.23%).The 4DVAR assimilation algorithm effectively improved the yield estimation accuracy of winter wheat at a regional scale.

关 键 词：冬小麦 估产 四维变分 集合卡尔曼滤波 Sentinel多源数据 CERESWheat模型 

分 类 号：S127[农业科学—农业基础科学] S512.1+1